Axioprax LabSynthetic DataNot Client Work

LAB-001
Lead-to-Revenue Engine

A reference implementation showing how inbound leads can be researched, qualified, routed, followed up, and monitored automatically. From web form submission to CRM record, first response, escalation, and reporting.

Pipeline steps
12
End-to-end workflow nodes
Systems connected
6
n8n, HubSpot, Claude AI, PostgreSQL, Email, Slack
Failure scenarios
7
Each with detection, response, and outcome
API calls per lead
~6
Enrichment, AI, CRM, email, Slack, log
Demo target
<10s
End-to-end in synthetic environment. Actual production time varies.
Estimated AI cost
<$0.05
Per standard demo execution at Haiku pricing. Varies in production.

Synthetic environment, not client production metrics. Actual production performance and cost vary based on models, APIs, retry behaviour, data volume, and workflow complexity. Demo target and estimated cost are design goals for the reference implementation, not guaranteed production metrics.

Transparency

Reference implementation using synthetic data only. The modelled company is Northstar B2B Systems (fictional). No real client data, client names, or real performance results are presented. All metrics are modelled with stated assumptions. Modelled scenario: Northstar B2B Systems receives ~200 inbound leads per month. Sales team has no automation.

Before vs After

What changes after the automation

Lead capture

Form submitted → email lands in the sales inbox → someone checks it later that day, or the next morning.

Research

Sales rep googles the company, checks LinkedIn, and pastes scattered notes into a spreadsheet before CRM entry.

CRM entry

Rep manually creates a contact and company record in HubSpot. Fields are incomplete and data is inconsistent across reps.

Assignment

Sales manager manually decides who gets the lead, usually based on memory of who owns which territory.

First response

Average first response: 4 to 8 hours. Often the next business day. Qualified leads cool off in the meantime.

Follow-up

Depends entirely on rep discipline. Some follow up three times. Some forget. No consistency across the team.

Visibility

Manager has no real-time view. Weekly CRM check. No alerting. Missed leads discovered in retrospect.

System Architecture

How the system is structured

Eight layers from input to reporting. Click any layer to see what it does and why it is designed that way.

Reliability

What happens when things go wrong

Most automation demos show the happy path. This shows the failure paths: what each system does when data is invalid, APIs are unavailable, or the AI returns an unexpected result.

Pick a failure and watch the run replay: where it breaks, and how the system recovers or hands over to a person.

Injected failure

Invalid lead data

  1. Arrive
  2. Validate
  3. FAIL (failure)
  4. Review Queue
  5. Human Correction
Detection
JSON schema validation on arrival, required fields, type checks, and format validation.
Response
Lead enters a human review queue. The workflow halts cleanly at this step.
Outcome
Ops team reviews and corrects. No data is silently dropped or passed to downstream systems.
Human Review

Where humans stay in the loop

Automation does not mean removing human judgment. These are the points where the system deliberately holds for review.

Low-confidence qualification

When AI confidence falls below the defined threshold, the lead is held for human review before any outreach is sent.

High-value accounts

Leads above a defined company size or deal-value threshold require manager approval before automated outreach begins.

Ambiguous AI draft

If the AI-drafted email is flagged for tone or compliance-sensitive language, it is held for human review.

Unclear territory

When territory rules produce no clear match, assignment goes to the sales manager for manual routing.

Exception queue entries

Any lead that fails validation, deduplication, or enrichment past the retry limit lands in a human review queue. Not the bin.

Technical Proof

Reference artifacts

Synthetic representations of what the system produces. All company and person data is fictional.

n8n: Enrichment Node (synthetic)Representative Synthetic Artifact
{
  "node": "HTTP Request",
  "name": "Enrich Lead",
  "parameters": {
    "method": "GET",
    "url": "https://enrichment.example/v1/company",
    "qs": {
      "domain": "={{ $json.email.split('@')[1] }}"
    },
    "retryOnFail": true,
    "maxTries": 3,
    "waitBetweenTries": 2000
  },
  "onError": "continueErrorOutput"
}
ROI Model

Put your own numbers in

Illustrative model only. Adjust inputs to match your context. Not client data.

200
8 min
5 min
$40/hr

Monthly manual hours

200 leads × 13 min ÷ 60

43h

Hours potentially recovered

85% automation assumption

~37h / mo

Monthly labour value

At $40/hr

$1,480

Estimated annual value

Illustrative only · not client results

$17,760
Assumptions: 85% manual effort eliminated · 200 leads/month · 13 min per lead · $40/hr · Illustrative model only, not client results

A person reads every brief and replies with questions or next steps.